A Grey-box Modelling and Its Application in Model-based Fault Detection

In order to provide an accurate and robust model with model-based fault detection, this paper combines a mathematical model and neural networks to develop a grey-box model. In the grey-box model, the mathematical model represents the dominant behaviour of the system, leaving the mismatch part of the system to be approximated by neural networks. The output of the grey-box model is used for residual generation in the model-based fault detection approach. Because the neural network compensates the model error from the mathematical model, a high accuracy model can be obtained and the residual generated under normal conditions can also be minimised by the combination. On the other hand, because most of the mathematical model mismatches exist in transients, the working load of the neural network can be reduced and the network structure can be simplified by the combination. Moreover, the grey box model provides more robust residual than black-box model and it enables the residual signatures to be physically interpretable. The capability of this grey-box model-based approach is evaluated in model accuracy and sensitivity in detecting faults introduced on an electro-hydraulic control system